Rod models in continuum and soft robot control: a review

Continuum and soft robots can transform diverse sectors, including healthcare, agriculture, marine, and space, thanks to their potential to adaptively interact with unstructured environments. These robots exhibit complex mechanics that pose diverse challenges in modeling and control. Among various models, continuum mechanical models based on rod theories can effectively capture the deformations of slender bodies in contact-rich scenarios. This structured review paper focuses on the role of rod models in continuum and soft robot control with a vertical approach. We provide a comprehensive summary of the mathematical background underlying the four main rod theories applied in soft robotics and their variants. Then, we review the literature on rod models applied to continuum and soft robots, providing a novel categorization in deformation classes. Finally, we survey recent model-based and learning-based control strategies leveraging rod models, highlighting their potential in real-world manipulation. We critically discuss the trends, advantages, limitations, research gaps, and possible future developments of rod models. This paper aims to guide researchers who intend to simulate and control new soft robots while providing feedback to the design and manufacturing community.

💡 Research Summary

This paper presents a comprehensive, vertically‑oriented review of rod‑theoretic models and their use in the modeling, simulation, and control of continuum and soft robots. The authors begin by motivating the need for rod models: continuum and soft robots possess an infinite‑dimensional configuration space, large nonlinear deformations, and frequent contact with unstructured environments, which render classic rigid‑body or simple geometric models insufficient. Four principal rod theories—Cosserat‑Reissner, Kirchhoff‑Love, Timoshenko, and Euler‑Bernoulli—are introduced, each with its underlying assumptions, governing equations, and discretization strategies (finite‑difference, finite‑element, and reduced‑order approximations). The review systematically compares these theories: Cosserat‑Reissner captures shear and torsion at the cost of higher computational load; Kirchhoff‑Love neglects shear, offering faster real‑time performance but reduced accuracy under strong shear; Timoshenko adds a shear‑correction term, providing a middle ground; Euler‑Bernoulli assumes pure bending, suitable only for slender, low‑shear scenarios.

A novel taxonomy is proposed that classifies existing rod‑based works into nine deformation classes (pure bending, torsion, axial stretch/compression, shear, coupled bending‑shear, bending‑torsion, etc.). For each class the authors list the adopted rod theory, actuation layout (tendon‑driven, pneumatic, dielectric, magnetic), material model (hyperelastic, viscoelastic, hysteretic), and experimental validation method. This classification highlights how the same mechanical effect can be captured by different rod formulations, guiding researchers to select the most appropriate model for a given task.

The control portion of the review is divided into model‑based and learning‑based strategies. Model‑based methods include inverse kinematics/dynamics, feedback linearization, model‑predictive control (MPC), optimal control, and robust/adaptive schemes that explicitly compensate for parameter uncertainties and external disturbances. The authors discuss how the structure of the rod equations (e.g., PDE vs. ODE after discretization) influences controller design, computational latency, and stability guarantees. In the learning‑based domain, reinforcement learning (RL) is emphasized as a powerful tool for contact‑rich manipulation. The paper surveys model‑based RL (using the rod model as a simulator for policy learning), model‑free RL with domain randomization, and hybrid approaches that fuse analytical rod dynamics with neural network residuals. Empirical results from recent works demonstrate that RL policies trained on accurate rod simulators can achieve higher success rates in tasks such as object grasping, tissue manipulation, and underwater locomotion.

Finally, the authors critically assess current limitations: (1) difficulty in modeling complex, multi‑material soft structures with strong non‑linearities (e.g., hysteresis, stress‑softening); (2) computational bottlenecks that hinder real‑time control for high‑fidelity rod models; (3) challenges in scaling rod‑based simulations to large‑scale contact scenarios; and (4) scarcity of standardized benchmarks for validation. They propose future research directions, including multi‑physics and multi‑scale integration, data‑driven parameter identification, hardware‑in‑the‑loop testing platforms, and the development of reduced‑order yet physically consistent rod models.

In summary, the paper serves as a detailed roadmap for researchers aiming to employ rod theories in the design, simulation, and control of continuum and soft robots. By unifying mathematical foundations, categorizing deformation‑based literature, and juxtaposing model‑based and learning‑based control techniques, it offers actionable guidance for advancing the state‑of‑the‑art in soft robot manipulation and interaction.